Sep 9, 2026
ACEWORKS Shares Achievements in AI-Based Business Automation and Vision Model
From "enhancing internal work efficiency using generative AI" to "automating the development process of autonomous driving vision models"... These are the main achievements of the internship projects shared by interns Jung Yeo-jin and Yang Seo-yun at the final overseas internship presentation held by ACEWORKS on the 28th of last month.
What is interesting is that the two interns did not stop at simply learning technologies, but focused on directly defining problems that repeatedly occur in the field and improving them using AI and software technology.
First, Jung Yeo-jin carried out a project under the theme of "AI Development and Application Based on Document Knowledge," structuring various documents accumulated within the company into knowledge that AI can utilize and applying it to actual work. Starting with an HR and internal knowledge search chatbot based on standard employment rules and ACEWORKS internal regulations, she verified the possibility of internal information exploration by AI by building a technical document search system using LiDAR brochures and manuals.
In particular, she expanded the project scope to sales operations, developing a monitoring system that allows AI to analyze Nara Marketplace bidding notices and task instructions to select candidate business projects that ACEWORKS can execute. She also confirmed the potential to reduce repetitive search and review tasks by comparing task requirements with existing proposal contents to review feasibility and even support drafting proposals.
Yang Seo-yun developed a "Vision Model Fine-Tuning Automation Dashboard." Autonomous driving vision models require additional training when new signals or objects not included in existing training data appear, a process that demands a significant amount of manual work such as data selection, labeling, training data composition, and iterative training. To reduce the workload of this fine-tuning process and increase accessibility, Yang integrated the related tasks into a single dashboard.
The developed dashboard consists of a 9-step pipeline ranging from data input, labeling, composition, training and validation data split, rehearsal data selection, data augmentation, training, to training result analysis. By automatically generating baseline labels using existing training models so users can focus on inspection and correction, and by allowing multiple training conditions to be executed at once with completion notifications, the dashboard reduced the burden of repetitive tasks. Another key feature is that the results generated at each step can be reused, minimizing rework.